Digital Twin Execution Report

The Digital Twin(DT) Execution Report shows the results of a DT execution.

Preparation

Python libraries are imported, and historical and simulation data is imported for evaluation.

Dependecies

%load_ext autotime

time: 212 µs

import pandas as pd
import plotly.express as px
import json
import os
import plotly.offline as pyo
pyo.init_notebook_mode()

time: 607 ms

import plotly.io as pio
#pio.renderers.default = "png"
png_renderer = pio.renderers["png"]
png_renderer.width = 4000
png_renderer.height = 2400

time: 430 µs

Load Data

ts = '2021-07-23 01:30:07.139913-'
base_path = "../../data/runs/" + ts

time: 278 µs

# Parameters
base_path = "/Users/georgekellerman/github/reflexer-digital-twin/data/runs/2021-07-23 02:02:05.954669-"

time: 263 µs

meta_path = base_path + "meta.json"
historical_path = base_path + "historical.csv.gz"
backtesting_path = base_path + "backtesting.csv.gz"
extrapolation_path = base_path + "extrapolation.csv.gz"

time: 364 µs

with open(meta_path, 'r') as fid:
    metadata = json.load(fid)
for i, row in metadata.items():
    print(f"{i}: {row}")
print("---")

createdAt: 2021-07-23 02:02:05.954669
initial_backtesting_timestamp: 2021-07-09 02:12:07
final_backtesting_timestamp: 2021-07-22 23:45:17
---
time: 1.77 ms

historical_df = pd.read_csv(historical_path).assign(origin='historical').iloc[1:].assign(subset=-1)
backtesting_df = pd.read_csv(backtesting_path).assign(origin='backtesting').iloc[1:]
extrapolation_df = pd.read_csv(extrapolation_path).assign(origin='extrapolation')

historical_df.loc[:, 'seconds_passed'] = backtesting_df.seconds_passed

past_df = (pd.concat([historical_df, backtesting_df])
             .reset_index(drop=True)
             .assign(seconds_passed=lambda df: df.seconds_passed - df.seconds_passed.min())          
          )

extrapolation_df.loc[:, 'seconds_passed'] += past_df.seconds_passed.max()
df = pd.concat([past_df, extrapolation_df])

df = (df
      .assign(hours_passed=lambda df: df.seconds_passed / (60 * 60))
      .assign(days_passed=lambda df: df.seconds_passed / (24 * 60 * 60))
     )

initial_time = pd.Timestamp(metadata['initial_backtesting_timestamp'])
deltas = df.hours_passed.map(lambda x: pd.Timedelta(x, unit='h'))
times = initial_time + deltas
df = df.assign(timestamp=times).reset_index()
last_time = df.query('origin == "extrapolation"').timestamp.min()

# Wrangling for extrapolation scenarios
def extrapolation_origin(_df):
    df = (_df.query("origin == 'extrapolation'")
             .assign(ewm_model=lambda df: df.ewm_model.fillna(0).astype(str))
             .assign(convergence_swap_intensity=lambda df: df.convergence_swap_intensity.fillna(0))
         )
    s = df.apply(lambda row: f"extrapolation (ewm={row.ewm_model}, csi={row.convergence_swap_intensity :.0%})", axis=1)
    return s

s = extrapolation_origin(df)
df.loc[s.index, 'origin'] = s

time: 37 ms

Visualizations

We will evaluate pricing feeds, the controller state, and the token state with visualizations.

Prices

value_cols = ('eth_price', 'market_price')
id_cols = {'timestamp', 'origin', 'subset'}
fig_df = (df.melt(id_vars=id_cols, value_vars=value_cols)
            .replace({'market_price': 'RAI Market Price (USD/RAI)'})
            .replace({'eth_price': 'ETH Price (USD/RAI)'})
         )
fig = px.line(fig_df,
              x='timestamp',
              y='value',
              color='origin',
              facet_row='variable',
              line_group='subset',
              labels={'market_price': 'RAI Market Price in USD'},
              height=600,
              title='Prices over time')
fig.update_traces(line=dict(width=0.5), 
                  marker=dict(opacity=0.05, size=5),
                  mode='lines+markers')
fig.add_vline(initial_time.timestamp() * 1000,
              annotation_text=initial_time.strftime('%Y-%m-%d %Hh'))
fig.add_vline(last_time.timestamp() * 1000,
              annotation_text=last_time.strftime('%Y-%m-%d %Hh'))
fig.add_vrect(x0=last_time, 
              x1=fig_df.timestamp.max(), 
              annotation_text="Extrapolation", 
              annotation_position="top right",
              fillcolor="cyan", 
              opacity=0.05)
fig.update_yaxes(matches=None)
fig.for_each_annotation(lambda a: a.update(text=a.text.split("=")[-1]))
fig.show()

time: 1.05 s

In the plots above we evaluate the Eth price data (upper) and the Rai market price data (lower) showing the historical data and the forward extrapolation under various assumptions Exponential Moving Average (EWM) and optimal swap assumptions. The CSI number indicating the maximum intensity of the 'optimal swap'.

Controller State

value_cols = ('redemption_price', 
              'redemption_rate', 
              'redemption_rate_annual',
              'proportional_error',
              'integral_error')
id_cols = {'timestamp', 'origin', 'subset'}

fig_df = (df.assign(redemption_rate_annual=lambda df: df.redemption_rate ** (24 * 365))
            .melt(id_vars=id_cols, value_vars=value_cols)
            .replace({'redemption_price': 'Redemption Price (USD/RAI)'})
            .replace({'redemption_rate': 'Redemption Rate (%/hours)'})
            .replace({'redemption_rate_annual': 'Redemption Rate (%/year)'})
            .replace({'proportional_error': 'Proportional Error (USD/RAI)'})
            .replace({'integral_error': 'Integral Error (USD * s / RAI)'})
         )
fig = px.line(fig_df,
              x='timestamp',
              y='value',
              color='origin',
              facet_row='variable',
              line_group='subset',
              height=1200,
              title='Controller State over time')
fig.update_traces(line=dict(width=0.5), 
                  marker=dict(opacity=0.05, size=5),
                  mode='lines+markers')
fig.add_vline(initial_time.timestamp() * 1000,
              annotation_text=initial_time.strftime('%Y-%m-%d %Hh'))
fig.add_vline(last_time.timestamp() * 1000,
              annotation_text=last_time.strftime('%Y-%m-%d %Hh'))
fig.add_vrect(x0=last_time, 
              x1=fig_df.timestamp.max(), 
              annotation_text="Extrapolation", 
              annotation_position="top right",
              fillcolor="cyan", 
              opacity=0.05)
fig.update_yaxes(matches=None)
fig.for_each_annotation(lambda a: a.update(text=a.text.split("=")[-1]))
fig.show()

time: 603 ms

In the plots above we evaluate the controller states, in descending order(i.e. first state is the first plot) of redemption_price, redemption_rate, redemption_rate_annual,proportional_error, and integral_error. The plots show the historical data and the forward extrapolation under various assumptions Exponential Moving Average (EWM) and optimal swap assumptions. The CSI number indicating the maximum intensity of the 'optimal swap'.

Token State

value_cols = ('rai_debt', 'eth_locked', 'rai_reserve', 'eth_reserve')
id_cols = {'timestamp', 'origin', 'subset'}
fig_df = (df.melt(id_vars=id_cols, value_vars=value_cols)
            .replace({'rai_debt': 'Global RAI Debt'})
            .replace({'eth_locked': 'ETH Collateral'})
            .replace({'rai_reserve': 'RAI reserve on Uniswap'})
            .replace({'eth_reserve': 'ETH reserve on Uniswap'})
         )
fig = px.line(fig_df,
              x='timestamp',
              y='value',
              color='origin',
              facet_row='variable',
              line_group='subset',
              labels={'market_price': 'RAI Market Price in USD'},
              height=900,
              title='Token State')
fig.update_traces(line=dict(width=0.5), 
                  marker=dict(opacity=0.05, size=5),
                  mode='lines+markers')
fig.add_vline(initial_time.timestamp() * 1000,
              annotation_text=initial_time.strftime('%Y-%m-%d %Hh'))
fig.add_vline(last_time.timestamp() * 1000,
              annotation_text=last_time.strftime('%Y-%m-%d %Hh'))
fig.add_vrect(x0=last_time, 
              x1=fig_df.timestamp.max(), 
              annotation_text="Extrapolation", 
              annotation_position="top right",
              fillcolor="cyan", 
              opacity=0.05)
fig.update_yaxes(matches=None)
fig.for_each_annotation(lambda a: a.update(text=a.text.split("=")[-1]))
fig.show()

time: 432 ms

In the plots above we evaluate the token states, in descending order(i.e. first state is the first plot), of: rai_debt, eth_locked, rai_reserve, and eth_reserve. The plots show the historical data and the forward extrapolation under various assumptions Exponential Moving Average (EWM) and optimal swap assumptions. The CSI number indicating the maximum intensity of the 'optimal swap'.

Conclusion

In this report, we've shown the real historical data that flow into the DT, and then extrapolations for the potential values of these states.